Previous commit 4df593496e broke the
"file" command on logical volume paths, since these are symbolic
links. We *should* follow these (only).
This inadvertantly broke virt-inspector too, which indicates that
we need more regression testing in this area. Since carrying whole
Fedora images around could make the distribution even larger than
now, I'm not sure at the moment how to do this.
Thanks to Matt Booth for diagnosing this bug.
The file call can hang if called on char devices (because we are
using the file -s option).
This is hard to solve cleanly without adding another file API.
However this restricts file to regular files, unless called explicitly
with a /dev/ path. For non-regular files, it will now return a
string like "directory".
There is a small semantic change for symbolic links. Previously
it would not have worked at all on absolute links (or rather, the
results would have been undefined). It would have treated relative
symlinks to regular files as the regular file itself. Now it will
return the string "symbolic link" in both cases.
This commit also makes the API safe when called on untrusted
filesystems. Previously a filesystem might have been set up so
that (eg) /etc/redhat-release was a char device, which would have
caused virt-inspector and virt-v2v to hang. Now it will not hang.
path = path to access file (/sysroot/.. or /dev/..)
display_path = original path, saved so we can display it
buf = optional buffer which is freed along return codepaths
There should be no change to the semantics of the code.
The guestfs_write call can be used to create small files with
arbitrary 8 bit content, including \0 bytes.
This replaces and deprecates write-file, which cannot be modified
to use BufferIn because of an unfortunate choice in the ABI: the
size parameter to write-file, if zero, means that the daemon tries
to calculate the length of the buffer using strlen. However this
fails if we pass a zero-length buffer using BufferIn because then
the daemon tries to do strlen on a (really) zero length buffer, not
even containing a terminating \0 character, thus segfaulting.
Pengzhen Cao noticed that read-file would fail for files
larger than the protocol size; this is *not* the bug. However
it would also lose protocol synchronization after this.
The reason was that functions which return RBufferOut in the
generator must not 'touch' the *size_r parameter along error
return paths.
I fixed read-file and initrd-cat, and I checked that pread was
doing the right thing.
This also adds regression tests for read-file with various categories
of large file.
chmod: Disallow negative mode, document mode affected by umask.
mkdir-mode: Disallow negative mode, document that filesystems
may interpret the mode in different ways.
mknod: Disallow negative mode, document mode affected by umask.
umask: Check the range of umask mask value carefully.
The RPC stubs already prefix the command name to error messages.
The daemon doesn't have to do this. As a (small) benefit this also
makes the daemon slightly smaller.
Code in the daemon such as:
if (argv[0] == NULL) {
reply_with_error ("passed an empty list");
return NULL;
}
now results in error messages like this:
><fs> command ""
libguestfs: error: command: passed an empty list
(whereas previously you would have seen ..command: command:..)
Instead of checking for futimens support and falling back
(incorrectly in one case) to using futimes, use gnulib's
module.
However the gnulib module does not yet support Win32, so
this change is only really useful on platforms like RHEL 5.
truncate, truncate_size: Used to truncate files to a particular
size, or to zero bytes.
mkdir_mode: Like mkdir but allows you to also specify the
initial permissions for the new directory.
utimens: Set timestamp on a file with nanosecond accuracy.
lchown: Corresponding to lchown(2) syscall (we already have chown).
The implementation is complicated by the fact that we had to
add an Int64 parameter type to the generator.
* src/generator.ml: Update all rules to handle Dev_or_Path.
(the above changes to generator.ml are mostly mechanical)
Emit a use of REQUIRE_ROOT_OR_RESOLVE_DEVICE.
* daemon/upload.c (do_download): Remove use of
REQUIRE_ROOT_OR_RESOLVE_DEVICE, now that it's automatically done
in calling code.
* daemon/file.c (do_file): Likewise.
Nearly every file-related function in daemons/*.c is affected:
Remove this pair of statements from each affected do_* function:
- NEED_ROOT (return -1);
- ABS_PATH (dir, return -1);
and change the type of the corresponding parameter to "const char *".
* src/generator.ml: Emit NEED_ROOT just once, even when there are two or
more Pathname args.
* daemon/file.c (do_file): Remove redundant use of ABS_PATH.
It's redundant because the preceding line invokes NEED_ROOT_OR_IS_DEVICE,
which also invokes ABS_PATH.
This commit implements the RBufferOut type for returning
arbitrary 8 bit data from calls.
We also implement the guestfs_read_file call to read a
whole file that can contain any 8 bit content, but up to
a limit of ~ 2 MB.
Currently /sysroot is hard-coded throughout the daemon code.
This patch turns the path into a variable so that we can change
it in future, for example to allow standalone mode to be implemented.
This patch was tested by running all the C API tests successfully.
